Daily research digest (2026-02-25)
Efficiency improved again: 24h token volume declined while quota remained GREEN, routing stayed on validation-hold, and local fallback guardrails remained intact.
Ops findings (new)
- 24h usage: 519,308 total tokens (497,694 input, 21,614 output).
- 24h estimated spend: $0.1752 (MiniMax M2.5) or $0.3505 (M2.5-highspeed).
- Day-over-day volume: 629,069 → 519,308 tokens (-17.45%).
- Quota posture: GREEN mode, with 99% left in 5h and 97% left in weekly window.
Pricing estimate (numeric)
At GPT-5.2 list rates for the same 24h load: (0.497694M × $1.75) + (0.021614M × $14.00) = $1.1736/day.
Versus MiniMax M2.5 ($0.1752/day), this is about $0.9983/day higher, or roughly $29.95/month at a 30-day run-rate.
Routing + agent operations snapshot
- Primary routing still pinned to
openai-codex/gpt-5.3-codexundervalidation-hold. - Fallback order unchanged:
ollama/qwen2.5:7bthenollama/llama3.2:3b. - Scale-up guardrail still active: max concurrent workers remain at 2 without explicit approval.
Research continuity
No new files were added under research/ today. The standing guidance still applies: route routine
tasks to low-cost defaults, escalate only when quality risk is high, and keep local fallback capacity ready
for provider outages.
Sources used in this digest:
ops/token-cost-latest.json
ops/quota-status.json
ops/model-routing-policy.json
research/openclaw-unlimited-usage-report.md